[ntp:questions] ntp server with two default routes misbehaving

Caecilius nospam at spamless.invalid
Mon Apr 21 09:21:12 UTC 2014


On Sun, 20 Apr 2014 02:27:04 GMT, "Jason Rabel"
<jason at extremeoverclocking.com> wrote:
>I can guess why it resets when things go out via a different IP & gateway. I think NTP keeps track of the relationships via Local IP
><-> Remote IP. So when your local IP is being forced to change because of the load balancing it breaks that relationship. 

I think that's exactly what's happening.

It looks like version 4.2.4p4 didn't care about the local IP, whereas
4.2.6p2 does.

>As a work-around I would simply configure all NTP traffic to go through one interface / route unless it fails. We are only talking
>about 128 byte packets, it's not going to put any load on one network link. Or possibly move the NTP server further in your network
>that is going to use one IP as its default gateway. In your case since the links are the same ISP I don't think there would be a
>latency mis-match.

Yes, I guess I could do something with iptables packet marking and
policy routing to force NTP traffic down one route.

For now, I've gone back to 4.2.4p4 while I assess the various options.

>While your situation isn't unique, I would imagine it would take a LOT of coding (and debugging) to resolve a situation like that,
>for only a very very very small handful of users. Most load balancers are on the edge of the network while NTP servers tend to be
>more internal.

I wonder why the change was made in the first place.  Did it improve
things for some users? It seems wrong to introduce something that
causes problems for some use cases unless it benefits others.

I wonder if it would cause problems for DHCP users, where the
ISP-allocated IP changes occasionally.  It probably wouldn't change
often enough to cause a big problem, but it would cause NTP to go
unsynchronised from time to time.

>Just my two cents...

Useful information, thanks.



More information about the questions mailing list